[Dovecot] litte problems upgrade from 1.0.x to 1.1.1

[Timo Sirainen]

On Jul 7, 2008, at 3:33 PM, Andre Hübner wrote:

>> Dovecot does not use mail messages to store folder meta-data as far  
>> as I know
>> so I highly doubt this message was created by Dovecot (does Dovecot  
>> even send
>> out messages except when going over quota?). Also notice the date of
>> 2006-09-18, which is almost two years in the past. The message was  
>> probably
>> created by some client and was somehow (willingly) ignored or  
>> flagged as
>> deleted. Could be that upgrading Dovecot to 1.1 made the message  
>> visible
>> again for whatever reason.
>> and
>> Anyway, this is most certainly not a Dovecot-internal message.
>>
>
> this case is really suspicious to me. don't hang on the old date  
> 2006-09-18. I can recreate it with dovecot 1.1.1 running on my  
> machine.
> I create completly new mbox, write to it and when i view mbox in  
> console i see just this mail. After deleting (pop3) this mail with  
> our webmailer or OutLook Express etc. I see this internal message  
> with current date.

Right. Dovecot creates such a "DO NOT DELETE" mail just as UW-IMAP/ 
Pine does. It's created when all messages from the mbox get deleted,  
so that UIDVALIDITY and NEXT-UID fields can be stored (without relying  
on them staying in index files).

But these messages should be ignored by Dovecot. If the first message  
of the mailbox has X-IMAP: header, it's skipped. I've really no idea  
why these would only now be visible with v1.1. Perhaps v1.0 had  
written buggy headers and v1.1 is more strict? I don't really know..  
Can you reproduce this in any way (so that the client even sees it)?

> unately i have now a more annoying error in imap processing.  
> Sometimes it happens that dovecot seems to be out of sync.
> My Log says:
>
> Jul  7 11:22:44 servername dovecot: IMAP(loginname): Next message  
> unexpectedly lost from 156756352

Looks like some kind of a locking problem.

> Jul  7 11:22:44 servername dovecot: child 21652 (imap) killed with  
> signal 11

Can you get a gdb backtrace from this crash? See http://dovecot.org/bugreport.html

> Jul  4 13:41:32 servername dovecot: IMAP(loginname): Next message  
> unexpectedly lost from 156715931
> Jul  4 13:41:32 servername dovecot: Panic: IMAP(loginname): file  
> message-parser.c: line 770 (message_parser_parse_next_block):  
> assertion failed: (ctx->input->eof
> || ctx->input->closed || ctx->input->stream_errno != 0 || ctx->broken)

Probably related to the message changing unexpectedly..

> lock_method = flock

See if it makes any difference to change this to fcntl.

> mbox_read_locks = dotlock
> mbox_write_locks = dotlock fcntl

How do you deliver new mails to the mbox?

> mbox_lazy_writes = no

Why? This makes performance worse.

> System is suse 10.1


What filesystem?
-------------- next part --------------
A non-text attachment was scrubbed...
Name: PGP.sig
Type: application/pgp-signature
Size: 194 bytes
Desc: This is a digitally signed message part
Url : http://dovecot.org/pipermail/dovecot/attachments/20080707/9ed740a6/attachment.bin